According to the official announcement, Google has officially launched the Gemini 3.8 Flash model, continuing its focus on extended software engineering and autonomous Agent capabilities. With a context length of approximately 1 million tokens and a maximum output of 65,000 tokens, Google-hosted Antigravity Agent has been automatically switched to this model. In internal benchmarks, Gemini 3.8 Flash secured first place in 8 out of 14 evaluations, with multiple capabilities approaching or surpassing those of Opus 5.

Specific results include DeepSWE v1.1 achieving 71.0% (Opus 5: 74.0%), Terminal-bench 2.1 reaching 89.4% (slightly above Opus 5's 89.1% and GPT-5.6 Sol's 88.8%), and attaining top scores in financial Agent, legal Agent, complex chart reasoning, and long-form video understanding tasks. However, it trailed behind Opus 5 on Terminal-bench 4.0 (19.1% vs. 51.8%) and OSWorld-2.0 (59.0% vs. 75.4%). Pricing for 3.8 Flash remains at $0.75 per million input tokens and $3.75 per million output tokens—consistent with 3.7 Flash—and represents only 15% of Opus 5’s cost. This promotional pricing is valid until December 31, after which rates will double starting January 1, 2027.